The answer is probably here: https://github.com/rurban/smhasher

Speed & number of cycles/hash is better with metro hash.

In my test metro hash gave me higher throughput, must be something with the xxhash implementation in Go

I tested both 64 bit variants a while ago and there was no throughput difference worth mentioning with the C++ and C implementations, respectively. (I settled on xxHash since it's around for longer and the code is portable, also I didn't need to do a C++->C translation).

If there is a large difference, perhaps one of the implementations is vectorized or botched.